Getting your client data into AI without the copy-paste
Every client has a reporting setup behind it. Data Studio dashboards that pull Google Ads, Meta, Google Analytics 4 (GA4), Search Console, Merchant Centre, Amazon, TikTok and Shopify into one place. When we need more detail than the dashboard shows, we go straight into each platform's own reports. Between them, we can answer most of the weekly questions: what did we spend, what did it bring back, where's the money going.
That covers reporting. It doesn't cover analysis.
Data Studio is very good at showing you a number for a particular date through formulas and blended data sources. It's no good at telling you why the number moved, or what to do next. For years that gap was filled by a person, a spreadsheet, and a spare couple of hours. You'd export the data, build a pivot table, stare at it, and write up what you found.
Artificial intelligence (AI) closed a lot of that gap. We can hand a much bigger dataset to Claude than anyone would happily work through by hand, and get a real read on it in minutes. Twelve months of orders. Every campaign across three ad platforms. A full year of email performance. The analysis that used to eat an afternoon now takes the length of a coffee.
There was one catch. Getting the data into the AI in the first place.
The slow bit was always the data
Until recently, feeding client data to Claude meant one of two things.
The first was downloading comma-separated values (CSV) files. A GA4 export here, a Shopify orders export there, Google Ads, Meta, Search Console, each in its own format with its own column names and its own date logic. You'd stitch them together, upload the lot, and start the analysis. By the time you'd finished, the numbers were already a day old. Do it weekly across five clients and you've lost a morning to admin before you've thought a single useful thought.
The second was pointing Claude at a browser and asking it to read the platform back-ends directly. We've done plenty of this, and it works, right up until it doesn't. Chrome and the ad platforms don't always get along. Logins drop. Two-factor prompts appear. Sessions time out mid-task. Tables load lazily, date pickers refuse to click, and a Shopify or GA4 screen that looks simple to a human is a maze to a browser agent. You end up babysitting the browser instead of analysing the data. Some days it's quicker than the CSV route. Some days it's slower and more annoying than doing the whole thing yourself.
Neither was the bottleneck we wanted. The thinking was fast. The fetching was slow.
MCP connectors changed that
Over the last few months, a better option has arrived: Model Context Protocol (MCP) connectors. In plain terms, an MCP connector is a direct, secure pipe between a tool that holds your data and the AI that wants to read it. No browser in the middle. No export. The AI asks the source for the data and the source hands it back.
This is our favourite: Power My Analytics (PMA).
We already used PMA, and it does a lot more than feed dashboards. It's a full data platform: it brings all your sources together (aggregation), cleans and standardises the data through proper extract, transform and load (ETL) work, and warehouses it in one place you can query. Feeding our Looker Studio dashboards is just one of the jobs we've had it doing for years, so the data plumbing was already in place before AI came along. It connects to more than 70 platforms across advertising, e-commerce, customer relationship management (CRM) and payments. What's new is that PMA now has a connector that gives Claude direct, read-only access to that same data.
So the export step disappears. Instead of pulling a CSV and uploading it, we ask Claude.
The feeds we run through PMA cover most of what a direct-to-consumer (DTC) brand needs watching: Google Ads, GA4, Meta across Facebook and Instagram, TikTok, Shopify, Klaviyo, Amazon Seller Central and Amazon Ads, plus Search Console. Set those up once and Claude can reach any of them on demand. Ask it to pull a metric, crunch the numbers, compare this month with last, flag anything that's moved too far, and draft the write-up. It works as the data gatherer and the first-pass analyst in one go. That's the part that's sped up our whole process.
Here are three ways we actually use it. The numbers in the examples are anonymised, but the workflows are real.
Example 1: scheduled weekly reports
Set it up once, and it runs every Monday without anyone remembering to start it.
Claude pulls last week's spend, revenue and return on ad spend (ROAS) across Google, Meta and Amazon for a client, compares it with the week before and the same week last year, writes a short commentary, and flags anything that moved more than a set threshold. It's done before we're at our desks.
Take an anonymised skincare account. The Monday report lands showing paid revenue up from roughly £2.45 to £3.57 back for every £1 spent week on week, with the gain driven by one Shopping campaign. Instead of starting the week by building that picture, we start it by reading it. A person checks the maths, adds the "so what", and it goes to the client the same morning.
The work that used to depend on someone having the time now happens on its own. We spend the freed-up hour on the decision, not the data pull.
Example 2: where your best customers actually come from
Most brands know their blended numbers. Fewer know whether the customers they acquire on Meta are worth as much as the ones from Google or email. That's a question worth answering before you decide where next quarter's budget goes.
We point Claude at the Shopify orders through PMA, split first-time orders from returning, and look at average basket size by acquisition channel. On an anonymised clothing client, the read came back clear: email-acquired first orders averaging around £58, paid social first orders nearer £41, with email also bringing a higher repeat rate inside 90 days. Same new customer, very different long-term value depending on where they walked in. Easy.
That used to be a pivot-table afternoon. Now it's a question we ask before lunch. The analysis is faster, which means we run it more often, which means the budget conversation is better informed.
Example 3: making sense of Amazon
Amazon is one of the more painful back-ends to export by hand. The data lives in different places, Seller Central and Amazon Ads don't sit in one tidy view, and joining the two by hand is a job nobody volunteers for.
With both feeds in PMA, we ask Claude to bring them together and answer the questions that matter: which products are still profitable after ad spend, where is advertising cost of sale creeping up, and how does the organic-to-paid split look across the range. On an anonymised account, the report surfaced three top sellers whose profit was being quietly eaten by rising ad costs, and a fourth that was doing most of its sales organically and barely needed the spend behind it.
Three exports and a manual join, replaced by one question to Claude. The platform that was hardest to analyse became one of the easiest.
The human check is the whole point
Read all that and it would be easy to think we've handed the job to a robot. We haven't, and we wouldn't, that’s not what this is about.
AI does the gathering and the first pass on the numbers. It does not sign off the report. Every dataset gets a human check before a client sees it, and that check does three things.
First, we verify the maths. AI is genuinely bad at arithmetic and will produce a confident, wrong percentage without blinking. We've caught plenty. The plausible-but-wrong number is the dangerous one, so every figure that reaches a client has been checked in a spreadsheet or against the source.
Second, we sense-check the read against what we already know about the account. A pattern that looks exciting in isolation often has a dull explanation: a promo, a stock issue, a tracking gap. We know these accounts. The AI doesn't.
Third, and most importantly, we write the recommendations ourselves. The data tells you what happened. Deciding what to do about it is where 28 years of running accounts earns its keep. We use AI to do more analysis on bigger datasets, faster. We spend the time it gives back on judgement, which is the part clients actually pay for.
That's the balance: AI at its most useful, doing the heavy lifting, with an experienced operator deciding what it means.
How to set it up
Two stages. First connect your data to PMA, then connect PMA to Claude. Exact button labels move around as the products update, but the process stays the same.
Connect your data sources to Power My Analytics
Create a Power My Analytics account, or log in to the one already feeding your dashboards.
Add each platform as a data source. You'll log in to each one and authorise read access: Google Ads, GA4, Meta, Shopify, Klaviyo, Amazon, and the rest. Repeat for every account you want covered.
Let it sync. The first sync backfills your history, so give it time before you expect a full picture.
Check each source is showing data before you move on.
Connect Power My Analytics to Claude
Open Settings in Claude. On the web app or the desktop app, click your profile initials at the bottom-left of the screen, then choose Settings.
Go to the Connectors tab. This is where every data connection Claude can use is listed.
Add the connector. If Power My Analytics shows up in the connector directory, select it. If not, choose "Add custom connector" and paste in the PMA connector link, which you'll find inside your Power My Analytics account under its integrations or MCP settings.
Authorise the connection. Claude opens a Power My Analytics sign-in window. Log in, then approve read access to your organisation's data. That authorisation step is what links your connected sources to Claude, and it's read-only, so Claude can see the data but can't change anything.
Check it's connected. Back in the Connectors tab, the connector should show as connected with your data sources listed underneath.
Test it. Open a new chat, make sure the Power My Analytics connector is switched on for that conversation, and ask something simple like last week's spend on one platform to confirm the pipe is live.
Once that's done, the export step is gone for good. You ask Claude, Claude then answers.
A few other things worth knowing about PMA
It's read-only. The connector reads your data, it doesn't touch your campaigns. Claude can't change a bid or pause an ad through it, which is exactly what you want when you're handing over access.
It scales past ads. More than 70 platforms, including CRM and payment tools, so the same setup stretches well beyond paid media if you want it to.
It does double duty. The same feeds still power your Data Studio and Google Sheets reporting. One connection, two uses: live dashboards for the client, direct access for your own analysis.
It keeps clients separate. Each organisation's accounts sit in their own space, so one client's data never bleeds into another's.
Keep an eye on connection health. Ad platforms expire logins on their own schedule, so a feed will occasionally need re-authorising. It's a small bit of housekeeping, and far less than a morning of CSV wrangling.
Worth an afternoon
If you run client reporting and the slow bit is wrangling the data rather than reading it, this is worth setting up. An afternoon to connect the sources, and the weekly admin that used to cost you mornings mostly disappears. The analysis gets faster, the reports get sharper, and the time you save goes back into the thinking.
If you'd rather we ran it for you, that's what our Strategy & Operations work is built for. We'll connect the data, build the reporting, and put an experienced operator between the numbers and your inbox. Get in touch and we'll show you what your data looks like once it's all in one place.
The connector itself is from Power My Analytics: powermyanalytics.com.
